A daily carbon emission prediction model combining two-stage feature selection and optimized extreme learning machine

Area of Science:

• Environmental Science
• Climate Change Modeling
• Data Science

Background:

• Global warming necessitates accurate carbon emission monitoring.
• Effective carbon emission reduction policies require precise, timely forecasting.
• Existing forecasting models may lack accuracy in capturing daily dynamics.

Purpose of the Study:

• To develop a novel daily carbon emission forecasting model.
• To enhance prediction accuracy through advanced decomposition and feature selection techniques.
• To provide a reliable tool for informing carbon emission reduction strategies.

Main Methods:

• Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN) for data decomposition.
• A two-stage feature selection using Partial Autocorrelation Function (PACF) and ReliefF.
• Extreme Learning Machine (ELM) optimized by Improved Sparrow Search Algorithm (ISSA) for prediction.

Main Results:

• The two-stage feature selection improved R², MAPE, and RMSE by 0.55%, 30.23%, and 28.46% respectively.
• The ISSA optimization further enhanced prediction accuracy, improving R², MAPE, and RMSE by 7.60%, 31.97%, and 44.79% respectively.
• The proposed model demonstrates significant improvements in forecasting accuracy.

Conclusions:

• The novel forecasting model effectively predicts daily carbon emissions.
• The combination of ICEEMDAN, two-stage feature selection, and ISSA-ELM offers superior accuracy.
• This model serves as a valuable reference for carbon emission reduction policies and future research.